Poster abstracts
Poster number 102 submitted by Myles Thornton-Sherman
TwisterML: Assessing the distribution and genomic context of twister ribozymes across the tree of life
Myles Thornton-Sherman (Graduate Program in Biochemistry, Microbiology, and Molecular Biology, Pennsylvania State University), Kobie Kirven (Graduate Program in Bioinformatics and Genomics, Pennsylvania State University), Philip Bevilacqua (Department of Chemistry, Pennsylvania State University)
Abstract:
Twister ribozymes are a widespread class of self-cleaving RNAs, yet their biological roles and distribution across the tree of life remain poorly understood. Although homology-based methods can identify twister candidates, they cannot distinguish intrinsically active ribozymes from inactive homologs without experimental characterization. To address this limitation, we developed TwisterML, a machine learning model that predicts intrinsic cleavage activity from sequence- and structure-derived features. Trained on the Cleavage High-Throughput Assay (CHiTA) dataset (1), TwisterML achieved an AUC-ROC of 0.85 ± 0.03 and, at a 5% false-positive rate, 94.5% specificity and 99.0% precision. We integrated TwisterML into a genome-scale pipeline and identified 1,841,675 twister candidates across almost 45,000 non-viral reference genomes, including 176,527 predicted intrinsically active ribozymes. Active twisters were overwhelmingly found in eukaryotes and showed strong lineage-specific enrichment, particularly among plants and several metazoan lineages. Twister abundance was positively associated with genome size but negatively associated with gene density, suggesting an association with genome architecture. Analysis of genomic context further revealed associations between twisters, specific Pfam protein-domain families, and transposable elements, including RTE-BovB LINEs. Together, these findings establish a scalable approach for identifying intrinsically active twister ribozymes and reveal phylogenetic and genomic associations that provide new hypotheses for their biological function and evolutionary history
References:
(1) McKinley LN, Bevilacqua PC. CHiTA: A scarless high-throughput pipeline for characterization of ribozymes. Methods. 2025;234:120–130.
Keywords: Ribozymes, Machine Learning